Vehicle trajectory reconstruction for freeway traffic considering lane changing behaviors

Abstract

Vehicle trajectory data provides critical information for many transportation applications. Due to limitations in the data collection techniques, usually, only partial trajectories can be obtained. As a result, trajectory reconstruction where the missing trajectories are inferenced by the observed data is an essential step for many downstream applications. Existing studies usually consider a connected vehicle (CV) environment for trajectory data collection and ignore the lane-changing (LC) behaviors in the reconstruction process. The deployment of connected and autonomous vehicles (CAVs) makes it possible to collect trajectory data more efficiently with much lower penetrations. This study proposes a vehicle trajectory reconstruction algorithm considering LC maneuvers in the CAV environment. The Pettit test and a rule-based optimization algorithm are designed to predict the possible LC time points. Then two car-following models are applied to reconstruct trajectories. The NGSIM US101 dataset is applied to evaluate the proposed reconstruction algorithm under varying CAV penetration rates (PRs) (e.g., 2%, 3%, 5%). The prediction of LC time points achieves high accuracy with average prediction errors less than 1 s under CAV PRs greater than 2%. Compared to the ground truth trajectories, the reconstructed trajectories have the mean absolute error (MAE) less than one vehicle length under 3% and higher CAV PRs.